Extracting of high quality biometric parameters from stationary image of handwritten signal

This work describes extracting of high quality biometric parameters from stationary image of handwritten signal. At first step, we perform binarization of initial handwritten signal images by Gabor filter and then map the pen movement directions. These maps are divided to segments, and pen movement direction histograms are plot for each segment. Those histograms are used as initial data during training of three-layer neural classifier by backpropagation algorithm. We propose to define high quality biometric parameters corresponding to data from the outputs of the neural classifier second layer. These extracted high quality biometric parameters can be used for training of neural transformer biometry-code according to methods described in the GOST R 52633.